Jeff King <peff@xxxxxxxx> writes:
> On Mon, Nov 21, 2016 at 10:15:43AM -0800, Junio C Hamano wrote:
>
>> > + test_cmp expect actual
>> > +'
>> > +
>>
>> Is this a recent regression? When applied on top of 'maint' or
>> older, it seems to pass just fine.
>>
>> ... Goes and looks ...
>>
>> Interesting. Peff's b9605bc4f2 ("config: only read .git/config from
>> configured repos", 2016-09-12) is where this starts failing, which
>> is understandable given the code change to builtin/stripspace.c in
>> [2/3].
>>
>> The analysis of the log message in [2/3] is wrong and needs
>> updating, though. In the old world order it worked by accident to
>> call git_config() without calling setup_git_directory(); after
>> b9605bc4f2, that no longer is valid and is exposed as a bug.
>
> Yeah, I noticed that while reading the patch. My b9605bc4f2 did regress
> this case, but called out the fact that "cd subdir && git stripspace"
> would never have worked. So one step back, 2 steps forward, and Dscho's
> patch is the right step forward.
Yes, absolutely.
I sent out a set of proposed amends, and the one for this step 1/3
runs the command inside a subdirectory to force it not to find the
.git/config file relative to its pwd, which can reveal the existing
breakage without help by b9605bc4f2 ;-) hence can be forked for
older maintenance tracks.